Fraud Score Explained: How to Evaluate Website Traffic Quality

Quick Answer

A fraud score is a number, usually 0 to 100, that estimates how risky an IP address, session, lead or transaction looks, where a higher number means higher risk. There is no universal standard. The same number means different things depending on the provider: MaxMind's score is a percentage probability that a transaction is fraudulent, Scamalytics describes its score as roughly the share of users seen from that IP who have been linked to fraud, and Stripe's score is a 0 to 99 rating with default cut-offs of 65 for elevated and 75 for high risk. To evaluate website traffic quality with a fraud score, score a sample of visits, group them by traffic source, compare each source with a trusted baseline, and set thresholds by testing scores against your own outcomes such as conversions, refunds and lead quality. Treat a high score as a reason to verify or investigate, not proof of fraud. If you buy traffic from adult or alternative ad networks, JuicyTraffic provides click fraud protection and ad tracking built for adult traffic.

Why Guides Give Different Answers

Search for what a good fraud score is and you will find thresholds that cannot all be right. One explainer says anything under 25 is clean and 50 or more should be rejected. Another says most legitimate users sit below 40 and many organisations treat 75 as elevated. Scamalytics's published bands run 0 to 19 low, 20 to 59 medium, 60 to 89 high and 90 to 100 very high. Others use 0 to 25, 26 to 60 and 61 to 100.

The differences have four causes:

Different providers, different scales. A 70 from one provider is not a 70 from another.
Different meanings. Some scores estimate the chance a transaction is fraudulent, others describe the reputation of an IP address or its network neighbourhood.
Thresholds are advice, not rules. Scamalytics itself says its bands are starting points that operators should tune to their own fraud data.
Most explainers are written for people checking their own IP, such as VPN and proxy users, not for someone evaluating where website traffic comes from.

What Different Scores Actually Mean

Score	Range	What it represents	Published guidance
MaxMind minFraud risk score	0.01 to 99	The percentage chance a transaction is fraudulent; 20 means a 20% chance	Set thresholds using the score distribution across users
MaxMind IP risk score	0.01 to 99	The chance a transaction is fraudulent based on the IP address alone	Interpret separately from the overall transaction score
Scamalytics fraud score	0 to 100	Approximately the share of users seen from that IP linked to fraud; 70 is roughly 7 in 10	Bands 0 to 19, 20 to 59, 60 to 89, 90 to 100; tune on your own data
IPQualityScore IP fraud score	0 to 100	A risk rating for an IP address	Many organisations treat above 75 as elevated risk; acceptable levels vary by business
Stripe Radar risk score	0 to 99	Payment risk from an adaptive model	By default 65 or above is elevated and 75 or above is high; adjustable for eligible accounts

An IP score and an overall transaction score answer different questions. Even if you pass only an IP address, do not assume that the overall score equals the IP score. Compare like-for-like signals and avoid treating a score of 50 as the midpoint of a normal traffic distribution.

What a Fraud Score Cannot Tell You

It rates a signal, not a person. A score based on an IP's reputation says something about its network neighbourhood, not about the session in front of you.
Shared IPs blur it. Google notes that offices, universities and mobile carriers share one public IP across many devices, so a risky IP can still carry real users.
A low score does not prove a human. A bot using a clean residential connection can look like an ordinary home user.
A high score does not prove fraud. It is a probability signal. Hard blocks should be reserved for a high score combined with strong behavioural evidence.
Scores drift. IP reputation changes as addresses are reassigned and abused, and MaxMind distinguishes between a snapshot of the past seven days and current risk.

How to Evaluate Website Traffic Quality With a Fraud Score

Step 1: Decide what you are scoring

Choose the unit: an IP at the click, a session, a lead, or a transaction. Different providers fit different units, so check what the score claims to measure before you use it.

Step 2: Collect source data

For each visit, capture the IP address, user agent, referrer, tracking parameter or sub-source ID, timestamp, and the outcome (conversion, refund, lead quality). IP addresses may count as personal data under privacy laws, so check your obligations before sending them to a third-party scoring service.

Step 3: Score a sample

Run a representative sample from each source through your scoring tool, not just the traffic you already suspect.

Step 4: Compare distributions by source

For each source, record the share of visits above your alert threshold, the median score, and the top flagged networks. Do not judge a source by its single worst visit. Compare every source with a trusted baseline, such as a verified customer cohort matched by geography and device. Branded search and direct traffic can also contain bots, so validate them before using them as a baseline.

Use a table like this:

Source	Visits sampled	Share above threshold	Median score	Conversion rate	Lead quality or refund rate
Baseline
Source A
Source B

Step 5: Calibrate thresholds with your own outcomes

This is the step most guides skip. Group visits into score buckets and compare what happened in each: conversion, refunds, chargebacks, fake or duplicate lead details. Scamalytics advises tuning thresholds on your own fraud data, and backtesting against your own outcomes helps you quantify the trade-off between catching abuse and challenging legitimate visitors. A stricter threshold catches more bad traffic but also blocks more real visitors.

Then set tiers:

Tier	Action
Low	Allow normally
Medium	Log and monitor, or add light friction
High	Challenge, rate-limit or send for review
Very high plus bad behaviour	Block

Step 6: Combine the score with behaviour and outcomes

A score is one signal. Pair it with interaction on the page, geography and device mix, sub-page visits, and what happens after the click. Our guides to buying website traffic and sophisticated invalid traffic cover behavioural checks in more detail.

Step 7: Act at the source first

If a source's flagged share is high compared with your baseline, pause or exclude the sub-source, ask for placement-level reporting, and request credits for invalid traffic. Blocking individual visitors is the last resort, not the first.

Step 8: Re-check regularly

Re-score samples monthly. Sources change, and so do IP reputations and provider models.

Common Mistakes

Treating one provider's bands as universal
Blocking on score alone, which can reject real users behind shared or carrier IPs
Using a score without a baseline, so you cannot tell what is normal for your audience
Never testing thresholds against conversions and refunds
Judging a source by one visit, not its distribution
Assuming a low score means human
Ignoring behaviour and post-click outcomes
Sending visitor data to a scoring service without checking privacy obligations

Checklist

Where JuicyTraffic Fits

A fraud score tells you how risky a signal looks. It does not tell you which of your ad networks is sending you the bots. For adult and alternative networks, where automated clicks and click farms are harder to spot and mainstream tools offer thinner coverage, JuicyTraffic provides click fraud protection and ad tracking built for adult traffic. It helps identify suspicious traffic and inform source exclusions, and gives you source-level tracking data so you can compare traffic quality across networks and placements.

FAQ

What is a fraud score? A number, usually from 0 to 100, estimating how risky an IP address, session, lead or transaction looks. Higher means riskier, but the exact meaning depends on the provider.

What is a good fraud score? Lower is better, but there is no universal cut-off. Published guidance ranges from "under 25 is clean" to "above 75 is elevated". Set your own thresholds by testing scores against your conversions, refunds and lead quality.

What fraud score gets traffic blocked? It depends on the site. Many sites challenge or block at high scores, but providers say thresholds are starting points. A score alone should not trigger a hard block without supporting evidence.

Is a fraud score a probability? Sometimes. MaxMind's score is a percentage chance a transaction is fraudulent. Scamalytics describes its score as roughly the share of users from an IP linked to fraud. Other scores are risk ratings with no stated probability.

Does a high fraud score prove fraud? No. It is a signal that should prompt verification, especially because shared IPs and VPN users can score high.

Does a low fraud score prove a visitor is real? No. Bots can use clean residential connections, so combine the score with behaviour and outcomes.

How do I use fraud scores to judge a traffic source? Score a sample of visits from each source, compare the share above your threshold with a baseline, and check whether high-score visits also convert worse or refund more.

What is the difference between an IP fraud score and a transaction fraud score? An IP score rates the reputation of an address. A transaction score rates a specific payment using many inputs, so the two should not be compared directly.

Should I send visitor IPs to a scoring service? IP addresses can be personal data under privacy laws, so check your legal obligations and the provider's terms first.

Final Take

A fraud score is a useful signal but a poor verdict. Know what your provider's number means, compare traffic sources with a baseline, calibrate thresholds on your own outcomes, and act at the source before you block visitors. If your traffic comes from adult or alternative ad networks, JuicyTraffic provides click fraud protection and ad tracking built for adult traffic, so more of your spend reaches real visitors.

Sources

About the Author

Dylan Dan is the founder of JuicyTraffic. He has spent 15 years specializing in adult advertising and ad-fraud prevention, helping advertisers assess traffic quality, identify invalid clicks and protect media budgets across dedicated ad networks.